Exploring The Brain Network Mechanism Of The Core Brain Regions Of Negative Symptoms Of Schizophrenia Based On Graph Theory And Graph Neural Network

Schizophrenia is a common neuropsychiatric disease.Previous studies have shown that the clinical characteristics of schizophrenia may be caused by abnormalities in brain structure and functional connectivity,which in turn affect changes in topological properties at the entire network level.We can use graph theory to study the large-scale network of the brain.A large-scale brain network refers to a complex network structure composed of many different brain regions.In recent years,studies have mostly explored the changes of network properties in schizophrenia from the global perspective of brain networks.Among them,the core nodes of large-scale brain networks play an important role in the information integration and transmission of the whole brain.At the same time,with the development of various machine learning methods,it has become more feasible to analyze complex brain imaging data,and network analysis based on graph convolution methods has become one of the hotspots for studying the characteristics of brain networks in schizophrenia.This paper mainly uses graph theory and graph convolutional neural network methods to explore the brain network mechanism of the core brain area of schizophrenia negative symptoms from the perspective of large-scale brain networks.Major studies include:(1)Based on the exploration of the core nodes of schizophrenia of the large-scale white matter brain network.In this paper,we first construct large-scale white matter brain networks using structural connectivity and augment structural brain networks with an indirect communication model.Secondly,using the Shortest Path Efficiency(SPE)module of schizophrenia patients and normal controls to divide the modules through the Louvain algorithm,and define the module division parameters by best matching with the inherent resting state network template.Subsequently,the participation coefficient of the brain area was calculated,and the change of the functional connection strength between modules in the two groups was calculated after the card threshold.At the same time,using the same module parameters,the above methods were used for schizophrenia patients and healthy people to screen brain regions with strong participation coefficients and large inter-group differences in topological organization.(2)Using graph convolutional neural network to classify schizophrenia.this study uses Magnetic Resonance Imaging(MRI)and takes large-scale brain network as the core analysis object,and proposes a method of using multimodal information fusion to classify schizophrenia.Graph Convolutional Deep Neural Network Algorithms for Objective Classification.First,this paper preprocesses theMRI images of multiple modalities,including structuralMRI,diffusionMRI and functionalMRI;then,the functional connection is used to extract multimodal image information on the Rich Club node calculated by the Harvard-Oxford template;finally,Treat brain regions as nodes,and use graph convolutional deep neural networks to classify nodes.To achieve the interpretability of the classification results,we apply the contrastive gradient-based saliency maps method to visualize the feature weights of the model.This study reveals the differences in neural networks between schizophrenia patients and healthy individuals through a graph theory-based analysis method.Through modularity analysis,it was found that functional connectivity among multiple brain regions was negatively correlated with Positive and Negative Syndrome Scale(PANSS)scores.Qualitative differences were investigated by calculating participation coefficients,identifying interacting nodes in the neural network,finding the left postcentral gyrus,left superior parietal lobule,and left superior lateral occipital cortex,compared with healthy controls,in schizophrenia patients in Subcortical gray matter exhibited widespread weakening of functional connectivity,further revealing abnormal connectivity patterns in brain networks in schizophrenia.The relationship between neural network connectivity and dysfunction was further explored by calculating the difference in FC profiles between schizophrenia patients and healthy individuals.The experimental results show that the research method in this paper can effectively classify schizophrenia patients and normal people.It is found that using the Rich Club node as the input feature can obtain the best classification effect and the precentral gyrus,postcentral gyrus,parallel lobular cortex and central tectum cortex was found to make a significant contribution to the classification results.In short,this paper found that the core nodes of large-scale brain networks play an important role in the information integration and transmission of the whole brain,providing new ideas for the diagnosis and treatment of schizophrenia.